DeepFace, a new face recognition technology with human level accuracy can be a boon or nightmare for privacy advocates.
We have seen face recognition technology in several Sci-Fi movies and it’s not new in the market. Even with the technology advancement, it is becoming cheaper, faster and easily available. A new dimension of possibilities has popped up with Facebook’s DeepFace. This technology can be boon for the security agencies but people anonymity won’t be possible.
The current face recognition system in Facebook (NASDAQ:FB) predicts whom users are trying to tag in photos, which, just like any computer, can’t be as accurate as expected. Lately, the company announced that it is working on a new technology called DeepFace that will operate it what they coined “near human accuracy” so that users won’t have to do it manually in the future.
DeepFace, a complex artificial intelligence system with facial detection features. It maps 3D facial features, creates a colorless model, and then narrows the specific characterization. The Facebook API group said that its accuracy method scored 97.25%, which is closely to human accuracy score of 97.5%.
While the concept of face recognition system isn’t new in the Facebook tagging options, the researchers said in the released report that, “In modern face recognition, the conventional pipeline consists of four stages: detect => align => represent => classify. We revisit both the alignment step and the representation step by employing explicit 3D face modeling in order to apply a piecewise affine transformation, and derive a face representation from a nine-layer deep neural network.”
To help them develop this technology, the company looked at 4.4 million tagged faces from the 4,030 users in its network to test the system. This will help it to better recognize the specific features of each user.
According to the researchers, “Thus we trained it on the largest facial dataset to-date, an identity labeled dataset of four million facial images belonging to more than 4,000 identities, where each identity has an average of over a thousand samples. The learned representations coupling the accurate model-based alignment with the large facial database generalize remarkably well to faces in unconstrained environments, even with a simple classifier.”
The company aims to bridge the gap between computer and human accuracy. However, we still have to see how the technology will run to strike the balance on privacy issues.